The concept is custom LLM ASICs, test link: https://chatjimmy.ai/ (Runs Llama 3.1 8B model as a proof-of-concept)